1 /*
2  * Copyright (c) 2000-2003 Silicon Graphics, Inc.  All Rights Reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it would be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11  *
12  * Further, this software is distributed without any warranty that it is
13  * free of the rightful claim of any third person regarding infringement
14  * or the like.  Any license provided herein, whether implied or
15  * otherwise, applies only to this software file.  Patent licenses, if
16  * any, provided herein do not apply to combinations of this program with
17  * other software, or any other product whatsoever.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, write the Free Software Foundation, Inc., 59
21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22  *
23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24  * Mountain View, CA  94043, or:
25  *
26  * http://www.sgi.com
27  *
28  * For further information regarding this notice, see:
29  *
30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31  */
32 
33 /*
34  * xfs_dir2_block.c
35  * XFS V2 directory implementation, single-block form.
36  * See xfs_dir2_block.h for the format.
37  */
38 
39 #include "xfs.h"
40 
41 #include "xfs_macros.h"
42 #include "xfs_types.h"
43 #include "xfs_inum.h"
44 #include "xfs_log.h"
45 #include "xfs_trans.h"
46 #include "xfs_sb.h"
47 #include "xfs_dir.h"
48 #include "xfs_dir2.h"
49 #include "xfs_dmapi.h"
50 #include "xfs_mount.h"
51 #include "xfs_bmap_btree.h"
52 #include "xfs_attr_sf.h"
53 #include "xfs_dir_sf.h"
54 #include "xfs_dir2_sf.h"
55 #include "xfs_dinode.h"
56 #include "xfs_inode_item.h"
57 #include "xfs_inode.h"
58 #include "xfs_da_btree.h"
59 #include "xfs_dir_leaf.h"
60 #include "xfs_dir2_data.h"
61 #include "xfs_dir2_leaf.h"
62 #include "xfs_dir2_block.h"
63 #include "xfs_dir2_trace.h"
64 #include "xfs_error.h"
65 
66 /*
67  * Local function prototypes.
68  */
69 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, xfs_dabuf_t *bp, int first,
70 				    int last);
71 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, xfs_dabuf_t *bp);
72 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, xfs_dabuf_t **bpp,
73 				     int *entno);
74 static int xfs_dir2_block_sort(const void *a, const void *b);
75 
76 /*
77  * Add an entry to a block directory.
78  */
79 int						/* error */
xfs_dir2_block_addname(xfs_da_args_t * args)80 xfs_dir2_block_addname(
81 	xfs_da_args_t		*args)		/* directory op arguments */
82 {
83 	xfs_dir2_data_free_t	*bf;		/* bestfree table in block */
84 	xfs_dir2_block_t	*block;		/* directory block structure */
85 	xfs_dir2_leaf_entry_t	*blp;		/* block leaf entries */
86 	xfs_dabuf_t		*bp;		/* buffer for block */
87 	xfs_dir2_block_tail_t	*btp;		/* block tail */
88 	int			compact;	/* need to compact leaf ents */
89 	xfs_dir2_data_entry_t	*dep;		/* block data entry */
90 	xfs_inode_t		*dp;		/* directory inode */
91 	xfs_dir2_data_unused_t	*dup;		/* block unused entry */
92 	int			error;		/* error return value */
93 	xfs_dir2_data_unused_t	*enddup=NULL;	/* unused at end of data */
94 	xfs_dahash_t		hash;		/* hash value of found entry */
95 	int			high;		/* high index for binary srch */
96 	int			highstale;	/* high stale index */
97 	int			lfloghigh=0;	/* last final leaf to log */
98 	int			lfloglow=0;	/* first final leaf to log */
99 	int			len;		/* length of the new entry */
100 	int			low;		/* low index for binary srch */
101 	int			lowstale;	/* low stale index */
102 	int			mid=0;		/* midpoint for binary srch */
103 	xfs_mount_t		*mp;		/* filesystem mount point */
104 	int			needlog;	/* need to log header */
105 	int			needscan;	/* need to rescan freespace */
106 	xfs_dir2_data_off_t	*tagp;		/* pointer to tag value */
107 	xfs_trans_t		*tp;		/* transaction structure */
108 
109 	xfs_dir2_trace_args("block_addname", args);
110 	dp = args->dp;
111 	tp = args->trans;
112 	mp = dp->i_mount;
113 	/*
114 	 * Read the (one and only) directory block into dabuf bp.
115 	 */
116 	if ((error =
117 	    xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
118 		return error;
119 	}
120 	ASSERT(bp != NULL);
121 	block = bp->data;
122 	/*
123 	 * Check the magic number, corrupted if wrong.
124 	 */
125 	if (unlikely(INT_GET(block->hdr.magic, ARCH_CONVERT)
126 						!= XFS_DIR2_BLOCK_MAGIC)) {
127 		XFS_CORRUPTION_ERROR("xfs_dir2_block_addname",
128 				     XFS_ERRLEVEL_LOW, mp, block);
129 		xfs_da_brelse(tp, bp);
130 		return XFS_ERROR(EFSCORRUPTED);
131 	}
132 	len = XFS_DIR2_DATA_ENTSIZE(args->namelen);
133 	/*
134 	 * Set up pointers to parts of the block.
135 	 */
136 	bf = block->hdr.bestfree;
137 	btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
138 	blp = XFS_DIR2_BLOCK_LEAF_P_ARCH(btp, ARCH_CONVERT);
139 	/*
140 	 * No stale entries?  Need space for entry and new leaf.
141 	 */
142 	if (INT_ISZERO(btp->stale, ARCH_CONVERT)) {
143 		/*
144 		 * Tag just before the first leaf entry.
145 		 */
146 		tagp = (xfs_dir2_data_off_t *)blp - 1;
147 		/*
148 		 * Data object just before the first leaf entry.
149 		 */
150 		enddup = (xfs_dir2_data_unused_t *)((char *)block + INT_GET(*tagp, ARCH_CONVERT));
151 		/*
152 		 * If it's not free then can't do this add without cleaning up:
153 		 * the space before the first leaf entry needs to be free so it
154 		 * can be expanded to hold the pointer to the new entry.
155 		 */
156 		if (INT_GET(enddup->freetag, ARCH_CONVERT) != XFS_DIR2_DATA_FREE_TAG)
157 			dup = enddup = NULL;
158 		/*
159 		 * Check out the biggest freespace and see if it's the same one.
160 		 */
161 		else {
162 			dup = (xfs_dir2_data_unused_t *)
163 			      ((char *)block + INT_GET(bf[0].offset, ARCH_CONVERT));
164 			if (dup == enddup) {
165 				/*
166 				 * It is the biggest freespace, is it too small
167 				 * to hold the new leaf too?
168 				 */
169 				if (INT_GET(dup->length, ARCH_CONVERT) < len + (uint)sizeof(*blp)) {
170 					/*
171 					 * Yes, we use the second-largest
172 					 * entry instead if it works.
173 					 */
174 					if (INT_GET(bf[1].length, ARCH_CONVERT) >= len)
175 						dup = (xfs_dir2_data_unused_t *)
176 						      ((char *)block +
177 						       INT_GET(bf[1].offset, ARCH_CONVERT));
178 					else
179 						dup = NULL;
180 				}
181 			} else {
182 				/*
183 				 * Not the same free entry,
184 				 * just check its length.
185 				 */
186 				if (INT_GET(dup->length, ARCH_CONVERT) < len) {
187 					dup = NULL;
188 				}
189 			}
190 		}
191 		compact = 0;
192 	}
193 	/*
194 	 * If there are stale entries we'll use one for the leaf.
195 	 * Is the biggest entry enough to avoid compaction?
196 	 */
197 	else if (INT_GET(bf[0].length, ARCH_CONVERT) >= len) {
198 		dup = (xfs_dir2_data_unused_t *)
199 		      ((char *)block + INT_GET(bf[0].offset, ARCH_CONVERT));
200 		compact = 0;
201 	}
202 	/*
203 	 * Will need to compact to make this work.
204 	 */
205 	else {
206 		/*
207 		 * Tag just before the first leaf entry.
208 		 */
209 		tagp = (xfs_dir2_data_off_t *)blp - 1;
210 		/*
211 		 * Data object just before the first leaf entry.
212 		 */
213 		dup = (xfs_dir2_data_unused_t *)((char *)block + INT_GET(*tagp, ARCH_CONVERT));
214 		/*
215 		 * If it's not free then the data will go where the
216 		 * leaf data starts now, if it works at all.
217 		 */
218 		if (INT_GET(dup->freetag, ARCH_CONVERT) == XFS_DIR2_DATA_FREE_TAG) {
219 			if (INT_GET(dup->length, ARCH_CONVERT) + (INT_GET(btp->stale, ARCH_CONVERT) - 1) *
220 			    (uint)sizeof(*blp) < len)
221 				dup = NULL;
222 		} else if ((INT_GET(btp->stale, ARCH_CONVERT) - 1) * (uint)sizeof(*blp) < len)
223 			dup = NULL;
224 		else
225 			dup = (xfs_dir2_data_unused_t *)blp;
226 		compact = 1;
227 	}
228 	/*
229 	 * If this isn't a real add, we're done with the buffer.
230 	 */
231 	if (args->justcheck)
232 		xfs_da_brelse(tp, bp);
233 	/*
234 	 * If we don't have space for the new entry & leaf ...
235 	 */
236 	if (!dup) {
237 		/*
238 		 * Not trying to actually do anything, or don't have
239 		 * a space reservation: return no-space.
240 		 */
241 		if (args->justcheck || args->total == 0)
242 			return XFS_ERROR(ENOSPC);
243 		/*
244 		 * Convert to the next larger format.
245 		 * Then add the new entry in that format.
246 		 */
247 		error = xfs_dir2_block_to_leaf(args, bp);
248 		xfs_da_buf_done(bp);
249 		if (error)
250 			return error;
251 		return xfs_dir2_leaf_addname(args);
252 	}
253 	/*
254 	 * Just checking, and it would work, so say so.
255 	 */
256 	if (args->justcheck)
257 		return 0;
258 	needlog = needscan = 0;
259 	/*
260 	 * If need to compact the leaf entries, do it now.
261 	 * Leave the highest-numbered stale entry stale.
262 	 * XXX should be the one closest to mid but mid is not yet computed.
263 	 */
264 	if (compact) {
265 		int	fromidx;		/* source leaf index */
266 		int	toidx;			/* target leaf index */
267 
268 		for (fromidx = toidx = INT_GET(btp->count, ARCH_CONVERT) - 1,
269 			highstale = lfloghigh = -1;
270 		     fromidx >= 0;
271 		     fromidx--) {
272 			if (INT_GET(blp[fromidx].address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR) {
273 				if (highstale == -1)
274 					highstale = toidx;
275 				else {
276 					if (lfloghigh == -1)
277 						lfloghigh = toidx;
278 					continue;
279 				}
280 			}
281 			if (fromidx < toidx)
282 				blp[toidx] = blp[fromidx];
283 			toidx--;
284 		}
285 		lfloglow = toidx + 1 - (INT_GET(btp->stale, ARCH_CONVERT) - 1);
286 		lfloghigh -= INT_GET(btp->stale, ARCH_CONVERT) - 1;
287 		INT_MOD(btp->count, ARCH_CONVERT, -(INT_GET(btp->stale, ARCH_CONVERT) - 1));
288 		xfs_dir2_data_make_free(tp, bp,
289 			(xfs_dir2_data_aoff_t)((char *)blp - (char *)block),
290 			(xfs_dir2_data_aoff_t)((INT_GET(btp->stale, ARCH_CONVERT) - 1) * sizeof(*blp)),
291 			&needlog, &needscan);
292 		blp += INT_GET(btp->stale, ARCH_CONVERT) - 1;
293 		INT_SET(btp->stale, ARCH_CONVERT, 1);
294 		/*
295 		 * If we now need to rebuild the bestfree map, do so.
296 		 * This needs to happen before the next call to use_free.
297 		 */
298 		if (needscan) {
299 			xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block,
300 				&needlog, NULL);
301 			needscan = 0;
302 		}
303 	}
304 	/*
305 	 * Set leaf logging boundaries to impossible state.
306 	 * For the no-stale case they're set explicitly.
307 	 */
308 	else if (INT_GET(btp->stale, ARCH_CONVERT)) {
309 		lfloglow = INT_GET(btp->count, ARCH_CONVERT);
310 		lfloghigh = -1;
311 	}
312 	/*
313 	 * Find the slot that's first lower than our hash value, -1 if none.
314 	 */
315 	for (low = 0, high = INT_GET(btp->count, ARCH_CONVERT) - 1; low <= high; ) {
316 		mid = (low + high) >> 1;
317 		if ((hash = INT_GET(blp[mid].hashval, ARCH_CONVERT)) == args->hashval)
318 			break;
319 		if (hash < args->hashval)
320 			low = mid + 1;
321 		else
322 			high = mid - 1;
323 	}
324 	while (mid >= 0 && INT_GET(blp[mid].hashval, ARCH_CONVERT) >= args->hashval) {
325 		mid--;
326 	}
327 	/*
328 	 * No stale entries, will use enddup space to hold new leaf.
329 	 */
330 	if (INT_ISZERO(btp->stale, ARCH_CONVERT)) {
331 		/*
332 		 * Mark the space needed for the new leaf entry, now in use.
333 		 */
334 		xfs_dir2_data_use_free(tp, bp, enddup,
335 			(xfs_dir2_data_aoff_t)
336 			((char *)enddup - (char *)block + INT_GET(enddup->length, ARCH_CONVERT) -
337 			 sizeof(*blp)),
338 			(xfs_dir2_data_aoff_t)sizeof(*blp),
339 			&needlog, &needscan);
340 		/*
341 		 * Update the tail (entry count).
342 		 */
343 		INT_MOD(btp->count, ARCH_CONVERT, +1);
344 		/*
345 		 * If we now need to rebuild the bestfree map, do so.
346 		 * This needs to happen before the next call to use_free.
347 		 */
348 		if (needscan) {
349 			xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block,
350 				&needlog, NULL);
351 			needscan = 0;
352 		}
353 		/*
354 		 * Adjust pointer to the first leaf entry, we're about to move
355 		 * the table up one to open up space for the new leaf entry.
356 		 * Then adjust our index to match.
357 		 */
358 		blp--;
359 		mid++;
360 		if (mid)
361 			memmove(blp, &blp[1], mid * sizeof(*blp));
362 		lfloglow = 0;
363 		lfloghigh = mid;
364 	}
365 	/*
366 	 * Use a stale leaf for our new entry.
367 	 */
368 	else {
369 		for (lowstale = mid;
370 		     lowstale >= 0 &&
371 			INT_GET(blp[lowstale].address, ARCH_CONVERT) != XFS_DIR2_NULL_DATAPTR;
372 		     lowstale--)
373 			continue;
374 		for (highstale = mid + 1;
375 		     highstale < INT_GET(btp->count, ARCH_CONVERT) &&
376 			INT_GET(blp[highstale].address, ARCH_CONVERT) != XFS_DIR2_NULL_DATAPTR &&
377 			(lowstale < 0 || mid - lowstale > highstale - mid);
378 		     highstale++)
379 			continue;
380 		/*
381 		 * Move entries toward the low-numbered stale entry.
382 		 */
383 		if (lowstale >= 0 &&
384 		    (highstale == INT_GET(btp->count, ARCH_CONVERT) ||
385 		     mid - lowstale <= highstale - mid)) {
386 			if (mid - lowstale)
387 				memmove(&blp[lowstale], &blp[lowstale + 1],
388 					(mid - lowstale) * sizeof(*blp));
389 			lfloglow = MIN(lowstale, lfloglow);
390 			lfloghigh = MAX(mid, lfloghigh);
391 		}
392 		/*
393 		 * Move entries toward the high-numbered stale entry.
394 		 */
395 		else {
396 			ASSERT(highstale < INT_GET(btp->count, ARCH_CONVERT));
397 			mid++;
398 			if (highstale - mid)
399 				memmove(&blp[mid + 1], &blp[mid],
400 					(highstale - mid) * sizeof(*blp));
401 			lfloglow = MIN(mid, lfloglow);
402 			lfloghigh = MAX(highstale, lfloghigh);
403 		}
404 		INT_MOD(btp->stale, ARCH_CONVERT, -1);
405 	}
406 	/*
407 	 * Point to the new data entry.
408 	 */
409 	dep = (xfs_dir2_data_entry_t *)dup;
410 	/*
411 	 * Fill in the leaf entry.
412 	 */
413 	INT_SET(blp[mid].hashval, ARCH_CONVERT, args->hashval);
414 	INT_SET(blp[mid].address, ARCH_CONVERT, XFS_DIR2_BYTE_TO_DATAPTR(mp, (char *)dep - (char *)block));
415 	xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
416 	/*
417 	 * Mark space for the data entry used.
418 	 */
419 	xfs_dir2_data_use_free(tp, bp, dup,
420 		(xfs_dir2_data_aoff_t)((char *)dup - (char *)block),
421 		(xfs_dir2_data_aoff_t)len, &needlog, &needscan);
422 	/*
423 	 * Create the new data entry.
424 	 */
425 	INT_SET(dep->inumber, ARCH_CONVERT, args->inumber);
426 	dep->namelen = args->namelen;
427 	memcpy(dep->name, args->name, args->namelen);
428 	tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
429 	INT_SET(*tagp, ARCH_CONVERT, (xfs_dir2_data_off_t)((char *)dep - (char *)block));
430 	/*
431 	 * Clean up the bestfree array and log the header, tail, and entry.
432 	 */
433 	if (needscan)
434 		xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog,
435 			NULL);
436 	if (needlog)
437 		xfs_dir2_data_log_header(tp, bp);
438 	xfs_dir2_block_log_tail(tp, bp);
439 	xfs_dir2_data_log_entry(tp, bp, dep);
440 	xfs_dir2_data_check(dp, bp);
441 	xfs_da_buf_done(bp);
442 	return 0;
443 }
444 
445 /*
446  * Readdir for block directories.
447  */
448 int						/* error */
xfs_dir2_block_getdents(xfs_trans_t * tp,xfs_inode_t * dp,uio_t * uio,int * eofp,xfs_dirent_t * dbp,xfs_dir2_put_t put)449 xfs_dir2_block_getdents(
450 	xfs_trans_t		*tp,		/* transaction (NULL) */
451 	xfs_inode_t		*dp,		/* incore inode */
452 	uio_t			*uio,		/* caller's buffer control */
453 	int			*eofp,		/* eof reached? (out) */
454 	xfs_dirent_t		*dbp,		/* caller's buffer */
455 	xfs_dir2_put_t		put)		/* abi's formatting function */
456 {
457 	xfs_dir2_block_t	*block;		/* directory block structure */
458 	xfs_dabuf_t		*bp;		/* buffer for block */
459 	xfs_dir2_block_tail_t	*btp;		/* block tail */
460 	xfs_dir2_data_entry_t	*dep;		/* block data entry */
461 	xfs_dir2_data_unused_t	*dup;		/* block unused entry */
462 	char			*endptr;	/* end of the data entries */
463 	int			error;		/* error return value */
464 	xfs_mount_t		*mp;		/* filesystem mount point */
465 	xfs_dir2_put_args_t	p;		/* arg package for put rtn */
466 	char			*ptr;		/* current data entry */
467 	int			wantoff;	/* starting block offset */
468 
469 	mp = dp->i_mount;
470 	/*
471 	 * If the block number in the offset is out of range, we're done.
472 	 */
473 	if (XFS_DIR2_DATAPTR_TO_DB(mp, uio->uio_offset) > mp->m_dirdatablk) {
474 		*eofp = 1;
475 		return 0;
476 	}
477 	/*
478 	 * Can't read the block, give up, else get dabuf in bp.
479 	 */
480 	if ((error =
481 	    xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
482 		return error;
483 	}
484 	ASSERT(bp != NULL);
485 	/*
486 	 * Extract the byte offset we start at from the seek pointer.
487 	 * We'll skip entries before this.
488 	 */
489 	wantoff = XFS_DIR2_DATAPTR_TO_OFF(mp, uio->uio_offset);
490 	block = bp->data;
491 	xfs_dir2_data_check(dp, bp);
492 	/*
493 	 * Set up values for the loop.
494 	 */
495 	btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
496 	ptr = (char *)block->u;
497 	endptr = (char *)XFS_DIR2_BLOCK_LEAF_P_ARCH(btp, ARCH_CONVERT);
498 	p.dbp = dbp;
499 	p.put = put;
500 	p.uio = uio;
501 	/*
502 	 * Loop over the data portion of the block.
503 	 * Each object is a real entry (dep) or an unused one (dup).
504 	 */
505 	while (ptr < endptr) {
506 		dup = (xfs_dir2_data_unused_t *)ptr;
507 		/*
508 		 * Unused, skip it.
509 		 */
510 		if (INT_GET(dup->freetag, ARCH_CONVERT) == XFS_DIR2_DATA_FREE_TAG) {
511 			ptr += INT_GET(dup->length, ARCH_CONVERT);
512 			continue;
513 		}
514 
515 		dep = (xfs_dir2_data_entry_t *)ptr;
516 
517 		/*
518 		 * Bump pointer for the next iteration.
519 		 */
520 		ptr += XFS_DIR2_DATA_ENTSIZE(dep->namelen);
521 		/*
522 		 * The entry is before the desired starting point, skip it.
523 		 */
524 		if ((char *)dep - (char *)block < wantoff)
525 			continue;
526 		/*
527 		 * Set up argument structure for put routine.
528 		 */
529 		p.namelen = dep->namelen;
530 
531 		p.cook = XFS_DIR2_DB_OFF_TO_DATAPTR(mp, mp->m_dirdatablk,
532 						    ptr - (char *)block);
533 		p.ino = INT_GET(dep->inumber, ARCH_CONVERT);
534 #if XFS_BIG_INUMS
535 		p.ino += mp->m_inoadd;
536 #endif
537 		p.name = (char *)dep->name;
538 
539 		/*
540 		 * Put the entry in the caller's buffer.
541 		 */
542 		error = p.put(&p);
543 
544 		/*
545 		 * If it didn't fit, set the final offset to here & return.
546 		 */
547 		if (!p.done) {
548 			uio->uio_offset =
549 				XFS_DIR2_DB_OFF_TO_DATAPTR(mp, mp->m_dirdatablk,
550 					(char *)dep - (char *)block);
551 			xfs_da_brelse(tp, bp);
552 			return error;
553 		}
554 	}
555 
556 	/*
557 	 * Reached the end of the block.
558 	 * Set the offset to a nonexistent block 1 and return.
559 	 */
560 	*eofp = 1;
561 
562 	uio->uio_offset =
563 		XFS_DIR2_DB_OFF_TO_DATAPTR(mp, mp->m_dirdatablk + 1, 0);
564 
565 	xfs_da_brelse(tp, bp);
566 
567 	return 0;
568 }
569 
570 /*
571  * Log leaf entries from the block.
572  */
573 static void
xfs_dir2_block_log_leaf(xfs_trans_t * tp,xfs_dabuf_t * bp,int first,int last)574 xfs_dir2_block_log_leaf(
575 	xfs_trans_t		*tp,		/* transaction structure */
576 	xfs_dabuf_t		*bp,		/* block buffer */
577 	int			first,		/* index of first logged leaf */
578 	int			last)		/* index of last logged leaf */
579 {
580 	xfs_dir2_block_t	*block;		/* directory block structure */
581 	xfs_dir2_leaf_entry_t	*blp;		/* block leaf entries */
582 	xfs_dir2_block_tail_t	*btp;		/* block tail */
583 	xfs_mount_t		*mp;		/* filesystem mount point */
584 
585 	mp = tp->t_mountp;
586 	block = bp->data;
587 	btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
588 	blp = XFS_DIR2_BLOCK_LEAF_P_ARCH(btp, ARCH_CONVERT);
589 	xfs_da_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)block),
590 		(uint)((char *)&blp[last + 1] - (char *)block - 1));
591 }
592 
593 /*
594  * Log the block tail.
595  */
596 static void
xfs_dir2_block_log_tail(xfs_trans_t * tp,xfs_dabuf_t * bp)597 xfs_dir2_block_log_tail(
598 	xfs_trans_t		*tp,		/* transaction structure */
599 	xfs_dabuf_t		*bp)		/* block buffer */
600 {
601 	xfs_dir2_block_t	*block;		/* directory block structure */
602 	xfs_dir2_block_tail_t	*btp;		/* block tail */
603 	xfs_mount_t		*mp;		/* filesystem mount point */
604 
605 	mp = tp->t_mountp;
606 	block = bp->data;
607 	btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
608 	xfs_da_log_buf(tp, bp, (uint)((char *)btp - (char *)block),
609 		(uint)((char *)(btp + 1) - (char *)block - 1));
610 }
611 
612 /*
613  * Look up an entry in the block.  This is the external routine,
614  * xfs_dir2_block_lookup_int does the real work.
615  */
616 int						/* error */
xfs_dir2_block_lookup(xfs_da_args_t * args)617 xfs_dir2_block_lookup(
618 	xfs_da_args_t		*args)		/* dir lookup arguments */
619 {
620 	xfs_dir2_block_t	*block;		/* block structure */
621 	xfs_dir2_leaf_entry_t	*blp;		/* block leaf entries */
622 	xfs_dabuf_t		*bp;		/* block buffer */
623 	xfs_dir2_block_tail_t	*btp;		/* block tail */
624 	xfs_dir2_data_entry_t	*dep;		/* block data entry */
625 	xfs_inode_t		*dp;		/* incore inode */
626 	int			ent;		/* entry index */
627 	int			error;		/* error return value */
628 	xfs_mount_t		*mp;		/* filesystem mount point */
629 
630 	xfs_dir2_trace_args("block_lookup", args);
631 	/*
632 	 * Get the buffer, look up the entry.
633 	 * If not found (ENOENT) then return, have no buffer.
634 	 */
635 	if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
636 		return error;
637 	dp = args->dp;
638 	mp = dp->i_mount;
639 	block = bp->data;
640 	xfs_dir2_data_check(dp, bp);
641 	btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
642 	blp = XFS_DIR2_BLOCK_LEAF_P_ARCH(btp, ARCH_CONVERT);
643 	/*
644 	 * Get the offset from the leaf entry, to point to the data.
645 	 */
646 	dep = (xfs_dir2_data_entry_t *)
647 	      ((char *)block + XFS_DIR2_DATAPTR_TO_OFF(mp, INT_GET(blp[ent].address, ARCH_CONVERT)));
648 	/*
649 	 * Fill in inode number, release the block.
650 	 */
651 	args->inumber = INT_GET(dep->inumber, ARCH_CONVERT);
652 	xfs_da_brelse(args->trans, bp);
653 	return XFS_ERROR(EEXIST);
654 }
655 
656 /*
657  * Internal block lookup routine.
658  */
659 static int					/* error */
xfs_dir2_block_lookup_int(xfs_da_args_t * args,xfs_dabuf_t ** bpp,int * entno)660 xfs_dir2_block_lookup_int(
661 	xfs_da_args_t		*args,		/* dir lookup arguments */
662 	xfs_dabuf_t		**bpp,		/* returned block buffer */
663 	int			*entno)		/* returned entry number */
664 {
665 	xfs_dir2_dataptr_t	addr;		/* data entry address */
666 	xfs_dir2_block_t	*block;		/* block structure */
667 	xfs_dir2_leaf_entry_t	*blp;		/* block leaf entries */
668 	xfs_dabuf_t		*bp;		/* block buffer */
669 	xfs_dir2_block_tail_t	*btp;		/* block tail */
670 	xfs_dir2_data_entry_t	*dep;		/* block data entry */
671 	xfs_inode_t		*dp;		/* incore inode */
672 	int			error;		/* error return value */
673 	xfs_dahash_t		hash;		/* found hash value */
674 	int			high;		/* binary search high index */
675 	int			low;		/* binary search low index */
676 	int			mid;		/* binary search current idx */
677 	xfs_mount_t		*mp;		/* filesystem mount point */
678 	xfs_trans_t		*tp;		/* transaction pointer */
679 
680 	dp = args->dp;
681 	tp = args->trans;
682 	mp = dp->i_mount;
683 	/*
684 	 * Read the buffer, return error if we can't get it.
685 	 */
686 	if ((error =
687 	    xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
688 		return error;
689 	}
690 	ASSERT(bp != NULL);
691 	block = bp->data;
692 	xfs_dir2_data_check(dp, bp);
693 	btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
694 	blp = XFS_DIR2_BLOCK_LEAF_P_ARCH(btp, ARCH_CONVERT);
695 	/*
696 	 * Loop doing a binary search for our hash value.
697 	 * Find our entry, ENOENT if it's not there.
698 	 */
699 	for (low = 0, high = INT_GET(btp->count, ARCH_CONVERT) - 1; ; ) {
700 		ASSERT(low <= high);
701 		mid = (low + high) >> 1;
702 		if ((hash = INT_GET(blp[mid].hashval, ARCH_CONVERT)) == args->hashval)
703 			break;
704 		if (hash < args->hashval)
705 			low = mid + 1;
706 		else
707 			high = mid - 1;
708 		if (low > high) {
709 			ASSERT(args->oknoent);
710 			xfs_da_brelse(tp, bp);
711 			return XFS_ERROR(ENOENT);
712 		}
713 	}
714 	/*
715 	 * Back up to the first one with the right hash value.
716 	 */
717 	while (mid > 0 && INT_GET(blp[mid - 1].hashval, ARCH_CONVERT) == args->hashval) {
718 		mid--;
719 	}
720 	/*
721 	 * Now loop forward through all the entries with the
722 	 * right hash value looking for our name.
723 	 */
724 	do {
725 		if ((addr = INT_GET(blp[mid].address, ARCH_CONVERT)) == XFS_DIR2_NULL_DATAPTR)
726 			continue;
727 		/*
728 		 * Get pointer to the entry from the leaf.
729 		 */
730 		dep = (xfs_dir2_data_entry_t *)
731 			((char *)block + XFS_DIR2_DATAPTR_TO_OFF(mp, addr));
732 		/*
733 		 * Compare, if it's right give back buffer & entry number.
734 		 */
735 		if (dep->namelen == args->namelen &&
736 		    dep->name[0] == args->name[0] &&
737 		    memcmp(dep->name, args->name, args->namelen) == 0) {
738 			*bpp = bp;
739 			*entno = mid;
740 			return 0;
741 		}
742 	} while (++mid < INT_GET(btp->count, ARCH_CONVERT) && INT_GET(blp[mid].hashval, ARCH_CONVERT) == hash);
743 	/*
744 	 * No match, release the buffer and return ENOENT.
745 	 */
746 	ASSERT(args->oknoent);
747 	xfs_da_brelse(tp, bp);
748 	return XFS_ERROR(ENOENT);
749 }
750 
751 /*
752  * Remove an entry from a block format directory.
753  * If that makes the block small enough to fit in shortform, transform it.
754  */
755 int						/* error */
xfs_dir2_block_removename(xfs_da_args_t * args)756 xfs_dir2_block_removename(
757 	xfs_da_args_t		*args)		/* directory operation args */
758 {
759 	xfs_dir2_block_t	*block;		/* block structure */
760 	xfs_dir2_leaf_entry_t	*blp;		/* block leaf pointer */
761 	xfs_dabuf_t		*bp;		/* block buffer */
762 	xfs_dir2_block_tail_t	*btp;		/* block tail */
763 	xfs_dir2_data_entry_t	*dep;		/* block data entry */
764 	xfs_inode_t		*dp;		/* incore inode */
765 	int			ent;		/* block leaf entry index */
766 	int			error;		/* error return value */
767 	xfs_mount_t		*mp;		/* filesystem mount point */
768 	int			needlog;	/* need to log block header */
769 	int			needscan;	/* need to fixup bestfree */
770 	xfs_dir2_sf_hdr_t	sfh;		/* shortform header */
771 	int			size;		/* shortform size */
772 	xfs_trans_t		*tp;		/* transaction pointer */
773 
774 	xfs_dir2_trace_args("block_removename", args);
775 	/*
776 	 * Look up the entry in the block.  Gets the buffer and entry index.
777 	 * It will always be there, the vnodeops level does a lookup first.
778 	 */
779 	if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
780 		return error;
781 	}
782 	dp = args->dp;
783 	tp = args->trans;
784 	mp = dp->i_mount;
785 	block = bp->data;
786 	btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
787 	blp = XFS_DIR2_BLOCK_LEAF_P_ARCH(btp, ARCH_CONVERT);
788 	/*
789 	 * Point to the data entry using the leaf entry.
790 	 */
791 	dep = (xfs_dir2_data_entry_t *)
792 	      ((char *)block + XFS_DIR2_DATAPTR_TO_OFF(mp, INT_GET(blp[ent].address, ARCH_CONVERT)));
793 	/*
794 	 * Mark the data entry's space free.
795 	 */
796 	needlog = needscan = 0;
797 	xfs_dir2_data_make_free(tp, bp,
798 		(xfs_dir2_data_aoff_t)((char *)dep - (char *)block),
799 		XFS_DIR2_DATA_ENTSIZE(dep->namelen), &needlog, &needscan);
800 	/*
801 	 * Fix up the block tail.
802 	 */
803 	INT_MOD(btp->stale, ARCH_CONVERT, +1);
804 	xfs_dir2_block_log_tail(tp, bp);
805 	/*
806 	 * Remove the leaf entry by marking it stale.
807 	 */
808 	INT_SET(blp[ent].address, ARCH_CONVERT, XFS_DIR2_NULL_DATAPTR);
809 	xfs_dir2_block_log_leaf(tp, bp, ent, ent);
810 	/*
811 	 * Fix up bestfree, log the header if necessary.
812 	 */
813 	if (needscan)
814 		xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog,
815 			NULL);
816 	if (needlog)
817 		xfs_dir2_data_log_header(tp, bp);
818 	xfs_dir2_data_check(dp, bp);
819 	/*
820 	 * See if the size as a shortform is good enough.
821 	 */
822 	if ((size = xfs_dir2_block_sfsize(dp, block, &sfh)) >
823 	    XFS_IFORK_DSIZE(dp)) {
824 		xfs_da_buf_done(bp);
825 		return 0;
826 	}
827 	/*
828 	 * If it works, do the conversion.
829 	 */
830 	return xfs_dir2_block_to_sf(args, bp, size, &sfh);
831 }
832 
833 /*
834  * Replace an entry in a V2 block directory.
835  * Change the inode number to the new value.
836  */
837 int						/* error */
xfs_dir2_block_replace(xfs_da_args_t * args)838 xfs_dir2_block_replace(
839 	xfs_da_args_t		*args)		/* directory operation args */
840 {
841 	xfs_dir2_block_t	*block;		/* block structure */
842 	xfs_dir2_leaf_entry_t	*blp;		/* block leaf entries */
843 	xfs_dabuf_t		*bp;		/* block buffer */
844 	xfs_dir2_block_tail_t	*btp;		/* block tail */
845 	xfs_dir2_data_entry_t	*dep;		/* block data entry */
846 	xfs_inode_t		*dp;		/* incore inode */
847 	int			ent;		/* leaf entry index */
848 	int			error;		/* error return value */
849 	xfs_mount_t		*mp;		/* filesystem mount point */
850 
851 	xfs_dir2_trace_args("block_replace", args);
852 	/*
853 	 * Lookup the entry in the directory.  Get buffer and entry index.
854 	 * This will always succeed since the caller has already done a lookup.
855 	 */
856 	if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
857 		return error;
858 	}
859 	dp = args->dp;
860 	mp = dp->i_mount;
861 	block = bp->data;
862 	btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
863 	blp = XFS_DIR2_BLOCK_LEAF_P_ARCH(btp, ARCH_CONVERT);
864 	/*
865 	 * Point to the data entry we need to change.
866 	 */
867 	dep = (xfs_dir2_data_entry_t *)
868 	      ((char *)block + XFS_DIR2_DATAPTR_TO_OFF(mp, INT_GET(blp[ent].address, ARCH_CONVERT)));
869 	ASSERT(INT_GET(dep->inumber, ARCH_CONVERT) != args->inumber);
870 	/*
871 	 * Change the inode number to the new value.
872 	 */
873 	INT_SET(dep->inumber, ARCH_CONVERT, args->inumber);
874 	xfs_dir2_data_log_entry(args->trans, bp, dep);
875 	xfs_dir2_data_check(dp, bp);
876 	xfs_da_buf_done(bp);
877 	return 0;
878 }
879 
880 /*
881  * Qsort comparison routine for the block leaf entries.
882  */
883 static int					/* sort order */
xfs_dir2_block_sort(const void * a,const void * b)884 xfs_dir2_block_sort(
885 	const void			*a,	/* first leaf entry */
886 	const void			*b)	/* second leaf entry */
887 {
888 	const xfs_dir2_leaf_entry_t	*la;	/* first leaf entry */
889 	const xfs_dir2_leaf_entry_t	*lb;	/* second leaf entry */
890 
891 	la = a;
892 	lb = b;
893 	return INT_GET(la->hashval, ARCH_CONVERT) < INT_GET(lb->hashval, ARCH_CONVERT) ? -1 :
894 		(INT_GET(la->hashval, ARCH_CONVERT) > INT_GET(lb->hashval, ARCH_CONVERT) ? 1 : 0);
895 }
896 
897 /*
898  * Convert a V2 leaf directory to a V2 block directory if possible.
899  */
900 int						/* error */
xfs_dir2_leaf_to_block(xfs_da_args_t * args,xfs_dabuf_t * lbp,xfs_dabuf_t * dbp)901 xfs_dir2_leaf_to_block(
902 	xfs_da_args_t		*args,		/* operation arguments */
903 	xfs_dabuf_t		*lbp,		/* leaf buffer */
904 	xfs_dabuf_t		*dbp)		/* data buffer */
905 {
906 	xfs_dir2_data_off_t	*bestsp;	/* leaf bests table */
907 	xfs_dir2_block_t	*block;		/* block structure */
908 	xfs_dir2_block_tail_t	*btp;		/* block tail */
909 	xfs_inode_t		*dp;		/* incore directory inode */
910 	xfs_dir2_data_unused_t	*dup;		/* unused data entry */
911 	int			error;		/* error return value */
912 	int			from;		/* leaf from index */
913 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
914 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
915 	xfs_dir2_leaf_tail_t	*ltp;		/* leaf tail structure */
916 	xfs_mount_t		*mp;		/* file system mount point */
917 	int			needlog;	/* need to log data header */
918 	int			needscan;	/* need to scan for bestfree */
919 	xfs_dir2_sf_hdr_t	sfh;		/* shortform header */
920 	int			size;		/* bytes used */
921 	xfs_dir2_data_off_t	*tagp;		/* end of entry (tag) */
922 	int			to;		/* block/leaf to index */
923 	xfs_trans_t		*tp;		/* transaction pointer */
924 
925 	xfs_dir2_trace_args_bb("leaf_to_block", args, lbp, dbp);
926 	dp = args->dp;
927 	tp = args->trans;
928 	mp = dp->i_mount;
929 	leaf = lbp->data;
930 	ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAF1_MAGIC);
931 	ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
932 	/*
933 	 * If there are data blocks other than the first one, take this
934 	 * opportunity to remove trailing empty data blocks that may have
935 	 * been left behind during no-space-reservation operations.
936 	 * These will show up in the leaf bests table.
937 	 */
938 	while (dp->i_d.di_size > mp->m_dirblksize) {
939 		bestsp = XFS_DIR2_LEAF_BESTS_P_ARCH(ltp, ARCH_CONVERT);
940 		if (INT_GET(bestsp[INT_GET(ltp->bestcount, ARCH_CONVERT) - 1], ARCH_CONVERT) ==
941 		    mp->m_dirblksize - (uint)sizeof(block->hdr)) {
942 			if ((error =
943 			    xfs_dir2_leaf_trim_data(args, lbp,
944 				    (xfs_dir2_db_t)(INT_GET(ltp->bestcount, ARCH_CONVERT) - 1))))
945 				goto out;
946 		} else {
947 			error = 0;
948 			goto out;
949 		}
950 	}
951 	/*
952 	 * Read the data block if we don't already have it, give up if it fails.
953 	 */
954 	if (dbp == NULL &&
955 	    (error = xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &dbp,
956 		    XFS_DATA_FORK))) {
957 		goto out;
958 	}
959 	block = dbp->data;
960 	ASSERT(INT_GET(block->hdr.magic, ARCH_CONVERT) == XFS_DIR2_DATA_MAGIC);
961 	/*
962 	 * Size of the "leaf" area in the block.
963 	 */
964 	size = (uint)sizeof(block->tail) +
965 	       (uint)sizeof(*lep) * (INT_GET(leaf->hdr.count, ARCH_CONVERT) - INT_GET(leaf->hdr.stale, ARCH_CONVERT));
966 	/*
967 	 * Look at the last data entry.
968 	 */
969 	tagp = (xfs_dir2_data_off_t *)((char *)block + mp->m_dirblksize) - 1;
970 	dup = (xfs_dir2_data_unused_t *)((char *)block + INT_GET(*tagp, ARCH_CONVERT));
971 	/*
972 	 * If it's not free or is too short we can't do it.
973 	 */
974 	if (INT_GET(dup->freetag, ARCH_CONVERT) != XFS_DIR2_DATA_FREE_TAG || INT_GET(dup->length, ARCH_CONVERT) < size) {
975 		error = 0;
976 		goto out;
977 	}
978 	/*
979 	 * Start converting it to block form.
980 	 */
981 	INT_SET(block->hdr.magic, ARCH_CONVERT, XFS_DIR2_BLOCK_MAGIC);
982 	needlog = 1;
983 	needscan = 0;
984 	/*
985 	 * Use up the space at the end of the block (blp/btp).
986 	 */
987 	xfs_dir2_data_use_free(tp, dbp, dup, mp->m_dirblksize - size, size,
988 		&needlog, &needscan);
989 	/*
990 	 * Initialize the block tail.
991 	 */
992 	btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
993 	INT_SET(btp->count, ARCH_CONVERT, INT_GET(leaf->hdr.count, ARCH_CONVERT) - INT_GET(leaf->hdr.stale, ARCH_CONVERT));
994 	INT_ZERO(btp->stale, ARCH_CONVERT);
995 	xfs_dir2_block_log_tail(tp, dbp);
996 	/*
997 	 * Initialize the block leaf area.  We compact out stale entries.
998 	 */
999 	lep = XFS_DIR2_BLOCK_LEAF_P_ARCH(btp, ARCH_CONVERT);
1000 	for (from = to = 0; from < INT_GET(leaf->hdr.count, ARCH_CONVERT); from++) {
1001 		if (INT_GET(leaf->ents[from].address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR)
1002 			continue;
1003 		lep[to++] = leaf->ents[from];
1004 	}
1005 	ASSERT(to == INT_GET(btp->count, ARCH_CONVERT));
1006 	xfs_dir2_block_log_leaf(tp, dbp, 0, INT_GET(btp->count, ARCH_CONVERT) - 1);
1007 	/*
1008 	 * Scan the bestfree if we need it and log the data block header.
1009 	 */
1010 	if (needscan)
1011 		xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog,
1012 			NULL);
1013 	if (needlog)
1014 		xfs_dir2_data_log_header(tp, dbp);
1015 	/*
1016 	 * Pitch the old leaf block.
1017 	 */
1018 	error = xfs_da_shrink_inode(args, mp->m_dirleafblk, lbp);
1019 	lbp = NULL;
1020 	if (error) {
1021 		goto out;
1022 	}
1023 	/*
1024 	 * Now see if the resulting block can be shrunken to shortform.
1025 	 */
1026 	if ((size = xfs_dir2_block_sfsize(dp, block, &sfh)) >
1027 	    XFS_IFORK_DSIZE(dp)) {
1028 		error = 0;
1029 		goto out;
1030 	}
1031 	return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1032 out:
1033 	if (lbp)
1034 		xfs_da_buf_done(lbp);
1035 	if (dbp)
1036 		xfs_da_buf_done(dbp);
1037 	return error;
1038 }
1039 
1040 /*
1041  * Convert the shortform directory to block form.
1042  */
1043 int						/* error */
xfs_dir2_sf_to_block(xfs_da_args_t * args)1044 xfs_dir2_sf_to_block(
1045 	xfs_da_args_t		*args)		/* operation arguments */
1046 {
1047 	xfs_dir2_db_t		blkno;		/* dir-relative block # (0) */
1048 	xfs_dir2_block_t	*block;		/* block structure */
1049 	xfs_dir2_leaf_entry_t	*blp;		/* block leaf entries */
1050 	xfs_dabuf_t		*bp;		/* block buffer */
1051 	xfs_dir2_block_tail_t	*btp;		/* block tail pointer */
1052 	char			*buf;		/* sf buffer */
1053 	int			buf_len;
1054 	xfs_dir2_data_entry_t	*dep;		/* data entry pointer */
1055 	xfs_inode_t		*dp;		/* incore directory inode */
1056 	int			dummy;		/* trash */
1057 	xfs_dir2_data_unused_t	*dup;		/* unused entry pointer */
1058 	int			endoffset;	/* end of data objects */
1059 	int			error;		/* error return value */
1060 	int			i;		/* index */
1061 	xfs_mount_t		*mp;		/* filesystem mount point */
1062 	int			needlog;	/* need to log block header */
1063 	int			needscan;	/* need to scan block freespc */
1064 	int			newoffset;	/* offset from current entry */
1065 	int			offset;		/* target block offset */
1066 	xfs_dir2_sf_entry_t	*sfep;		/* sf entry pointer */
1067 	xfs_dir2_sf_t		*sfp;		/* shortform structure */
1068 	xfs_dir2_data_off_t	*tagp;		/* end of data entry */
1069 	xfs_trans_t		*tp;		/* transaction pointer */
1070 
1071 	xfs_dir2_trace_args("sf_to_block", args);
1072 	dp = args->dp;
1073 	tp = args->trans;
1074 	mp = dp->i_mount;
1075 	ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
1076 	/*
1077 	 * Bomb out if the shortform directory is way too short.
1078 	 */
1079 	if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1080 		ASSERT(XFS_FORCED_SHUTDOWN(mp));
1081 		return XFS_ERROR(EIO);
1082 	}
1083 	ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
1084 	ASSERT(dp->i_df.if_u1.if_data != NULL);
1085 	sfp = (xfs_dir2_sf_t *)dp->i_df.if_u1.if_data;
1086 	ASSERT(dp->i_d.di_size >= XFS_DIR2_SF_HDR_SIZE(sfp->hdr.i8count));
1087 	/*
1088 	 * Copy the directory into the stack buffer.
1089 	 * Then pitch the incore inode data so we can make extents.
1090 	 */
1091 
1092 	buf_len = dp->i_df.if_bytes;
1093 	buf = kmem_alloc(dp->i_df.if_bytes, KM_SLEEP);
1094 
1095 	memcpy(buf, sfp, dp->i_df.if_bytes);
1096 	xfs_idata_realloc(dp, -dp->i_df.if_bytes, XFS_DATA_FORK);
1097 	dp->i_d.di_size = 0;
1098 	xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1099 	/*
1100 	 * Reset pointer - old sfp is gone.
1101 	 */
1102 	sfp = (xfs_dir2_sf_t *)buf;
1103 	/*
1104 	 * Add block 0 to the inode.
1105 	 */
1106 	error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1107 	if (error) {
1108 		kmem_free(buf, buf_len);
1109 		return error;
1110 	}
1111 	/*
1112 	 * Initialize the data block.
1113 	 */
1114 	error = xfs_dir2_data_init(args, blkno, &bp);
1115 	if (error) {
1116 		kmem_free(buf, buf_len);
1117 		return error;
1118 	}
1119 	block = bp->data;
1120 	INT_SET(block->hdr.magic, ARCH_CONVERT, XFS_DIR2_BLOCK_MAGIC);
1121 	/*
1122 	 * Compute size of block "tail" area.
1123 	 */
1124 	i = (uint)sizeof(*btp) +
1125 	    (INT_GET(sfp->hdr.count, ARCH_CONVERT) + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1126 	/*
1127 	 * The whole thing is initialized to free by the init routine.
1128 	 * Say we're using the leaf and tail area.
1129 	 */
1130 	dup = (xfs_dir2_data_unused_t *)block->u;
1131 	needlog = needscan = 0;
1132 	xfs_dir2_data_use_free(tp, bp, dup, mp->m_dirblksize - i, i, &needlog,
1133 		&needscan);
1134 	ASSERT(needscan == 0);
1135 	/*
1136 	 * Fill in the tail.
1137 	 */
1138 	btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
1139 	INT_SET(btp->count, ARCH_CONVERT, INT_GET(sfp->hdr.count, ARCH_CONVERT) + 2);	/* ., .. */
1140 	INT_ZERO(btp->stale, ARCH_CONVERT);
1141 	blp = XFS_DIR2_BLOCK_LEAF_P_ARCH(btp, ARCH_CONVERT);
1142 	endoffset = (uint)((char *)blp - (char *)block);
1143 	/*
1144 	 * Remove the freespace, we'll manage it.
1145 	 */
1146 	xfs_dir2_data_use_free(tp, bp, dup,
1147 		(xfs_dir2_data_aoff_t)((char *)dup - (char *)block),
1148 		INT_GET(dup->length, ARCH_CONVERT), &needlog, &needscan);
1149 	/*
1150 	 * Create entry for .
1151 	 */
1152 	dep = (xfs_dir2_data_entry_t *)
1153 	      ((char *)block + XFS_DIR2_DATA_DOT_OFFSET);
1154 	INT_SET(dep->inumber, ARCH_CONVERT, dp->i_ino);
1155 	dep->namelen = 1;
1156 	dep->name[0] = '.';
1157 	tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
1158 	INT_SET(*tagp, ARCH_CONVERT, (xfs_dir2_data_off_t)((char *)dep - (char *)block));
1159 	xfs_dir2_data_log_entry(tp, bp, dep);
1160 	INT_SET(blp[0].hashval, ARCH_CONVERT, xfs_dir_hash_dot);
1161 	INT_SET(blp[0].address, ARCH_CONVERT, XFS_DIR2_BYTE_TO_DATAPTR(mp, (char *)dep - (char *)block));
1162 	/*
1163 	 * Create entry for ..
1164 	 */
1165 	dep = (xfs_dir2_data_entry_t *)
1166 		((char *)block + XFS_DIR2_DATA_DOTDOT_OFFSET);
1167 	INT_SET(dep->inumber, ARCH_CONVERT, XFS_DIR2_SF_GET_INUMBER_ARCH(sfp, &sfp->hdr.parent, ARCH_CONVERT));
1168 	dep->namelen = 2;
1169 	dep->name[0] = dep->name[1] = '.';
1170 	tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
1171 	INT_SET(*tagp, ARCH_CONVERT, (xfs_dir2_data_off_t)((char *)dep - (char *)block));
1172 	xfs_dir2_data_log_entry(tp, bp, dep);
1173 	INT_SET(blp[1].hashval, ARCH_CONVERT, xfs_dir_hash_dotdot);
1174 	INT_SET(blp[1].address, ARCH_CONVERT, XFS_DIR2_BYTE_TO_DATAPTR(mp, (char *)dep - (char *)block));
1175 	offset = XFS_DIR2_DATA_FIRST_OFFSET;
1176 	/*
1177 	 * Loop over existing entries, stuff them in.
1178 	 */
1179 	if ((i = 0) == INT_GET(sfp->hdr.count, ARCH_CONVERT))
1180 		sfep = NULL;
1181 	else
1182 		sfep = XFS_DIR2_SF_FIRSTENTRY(sfp);
1183 	/*
1184 	 * Need to preserve the existing offset values in the sf directory.
1185 	 * Insert holes (unused entries) where necessary.
1186 	 */
1187 	while (offset < endoffset) {
1188 		/*
1189 		 * sfep is null when we reach the end of the list.
1190 		 */
1191 		if (sfep == NULL)
1192 			newoffset = endoffset;
1193 		else
1194 			newoffset = XFS_DIR2_SF_GET_OFFSET_ARCH(sfep, ARCH_CONVERT);
1195 		/*
1196 		 * There should be a hole here, make one.
1197 		 */
1198 		if (offset < newoffset) {
1199 			dup = (xfs_dir2_data_unused_t *)
1200 			      ((char *)block + offset);
1201 			INT_SET(dup->freetag, ARCH_CONVERT, XFS_DIR2_DATA_FREE_TAG);
1202 			INT_SET(dup->length, ARCH_CONVERT, newoffset - offset);
1203 			INT_SET(*XFS_DIR2_DATA_UNUSED_TAG_P_ARCH(dup, ARCH_CONVERT), ARCH_CONVERT,
1204 				(xfs_dir2_data_off_t)
1205 				((char *)dup - (char *)block));
1206 			xfs_dir2_data_log_unused(tp, bp, dup);
1207 			(void)xfs_dir2_data_freeinsert((xfs_dir2_data_t *)block,
1208 				dup, &dummy);
1209 			offset += INT_GET(dup->length, ARCH_CONVERT);
1210 			continue;
1211 		}
1212 		/*
1213 		 * Copy a real entry.
1214 		 */
1215 		dep = (xfs_dir2_data_entry_t *)((char *)block + newoffset);
1216 		INT_SET(dep->inumber, ARCH_CONVERT, XFS_DIR2_SF_GET_INUMBER_ARCH(sfp,
1217 				XFS_DIR2_SF_INUMBERP(sfep), ARCH_CONVERT));
1218 		dep->namelen = sfep->namelen;
1219 		memcpy(dep->name, sfep->name, dep->namelen);
1220 		tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
1221 		INT_SET(*tagp, ARCH_CONVERT, (xfs_dir2_data_off_t)((char *)dep - (char *)block));
1222 		xfs_dir2_data_log_entry(tp, bp, dep);
1223 		INT_SET(blp[2 + i].hashval, ARCH_CONVERT, xfs_da_hashname((char *)sfep->name, sfep->namelen));
1224 		INT_SET(blp[2 + i].address, ARCH_CONVERT, XFS_DIR2_BYTE_TO_DATAPTR(mp,
1225 						 (char *)dep - (char *)block));
1226 		offset = (int)((char *)(tagp + 1) - (char *)block);
1227 		if (++i == INT_GET(sfp->hdr.count, ARCH_CONVERT))
1228 			sfep = NULL;
1229 		else
1230 			sfep = XFS_DIR2_SF_NEXTENTRY(sfp, sfep);
1231 	}
1232 	/* Done with the temporary buffer */
1233 	kmem_free(buf, buf_len);
1234 	/*
1235 	 * Sort the leaf entries by hash value.
1236 	 */
1237 	qsort(blp, INT_GET(btp->count, ARCH_CONVERT), sizeof(*blp), xfs_dir2_block_sort);
1238 	/*
1239 	 * Log the leaf entry area and tail.
1240 	 * Already logged the header in data_init, ignore needlog.
1241 	 */
1242 	ASSERT(needscan == 0);
1243 	xfs_dir2_block_log_leaf(tp, bp, 0, INT_GET(btp->count, ARCH_CONVERT) - 1);
1244 	xfs_dir2_block_log_tail(tp, bp);
1245 	xfs_dir2_data_check(dp, bp);
1246 	xfs_da_buf_done(bp);
1247 	return 0;
1248 }
1249